U.S. patent application number 12/727496 was filed with the patent office on 2011-09-22 for system and method for generating enhanced density distribution in a three dimenional model of a structure for use in skeletal assessment using a limited number of two-dimensional views.
Invention is credited to Krishna Ramamurthi, Kevin Wilson.
Application Number | 20110231162 12/727496 |
Document ID | / |
Family ID | 44647906 |
Filed Date | 2011-09-22 |
United States Patent
Application |
20110231162 |
Kind Code |
A1 |
Ramamurthi; Krishna ; et
al. |
September 22, 2011 |
System and Method for Generating Enhanced Density Distribution in a
Three Dimenional Model of a Structure for Use in Skeletal
Assessment Using a Limited Number of Two-Dimensional Views
Abstract
A method of generating a density enhanced model of an object is
described. The method includes generating a customized a model of
an object using a pre-defined set of models in combination with at
least one projection image of the object, where the customized
model is formed of a plurality of volume elements including density
information. A density map is generated by relating a synthesized
projection image of the customized model to an actual projection
image of the object. Gains from the density map are back-projected
into the customized model to provide a density enhanced customized
model of the object. Because the density map is calculated using
information from the synthesized projection image in combination
with actual projection images of the structure, it has been shown
to provide spatial geometry and volumetric density results
comparable to those of QCT but with reduced patient exposure,
equipment cost and examination time.
Inventors: |
Ramamurthi; Krishna;
(Somerville, MA) ; Wilson; Kevin; (Acton,
MA) |
Family ID: |
44647906 |
Appl. No.: |
12/727496 |
Filed: |
March 19, 2010 |
Current U.S.
Class: |
703/1 ; 345/419;
703/11; 703/2 |
Current CPC
Class: |
G06T 2210/41 20130101;
G06T 2211/436 20130101; A61B 6/5247 20130101; G06T 19/20 20130101;
G06T 11/006 20130101; G16H 50/50 20180101; A61B 6/505 20130101 |
Class at
Publication: |
703/1 ; 345/419;
703/2; 703/11 |
International
Class: |
G06F 17/50 20060101
G06F017/50; G06F 17/10 20060101 G06F017/10; G06G 7/60 20060101
G06G007/60; G06T 15/00 20060101 G06T015/00 |
Claims
1. A method of generating a density enhanced model of an object
includes the steps of: generating a customized a model of an object
using a generic model of the object, the customized model
comprising a plurality of volume elements, each volume element
including density information; generating a density map using a
synthesized projection image of the customized model and a
projection image of the object; back-projecting gains from the
density map into the customized model to modify the density
information of each of the volume elements to provide a density
enhanced customized model of the object and thereby transform the
generic model of the object into an object specific model; and
performing at least one of storing the density enhanced customized
model of the object and displaying the density enhanced customized
model of the object on a user display.
2. The method according to claim 1, wherein the object is a bone,
the density information is associated with a density of the bone,
and wherein the method includes the step of determining a bone mass
density (BMD) of the object using the density enhanced customized
model of the object.
3. The method according to claim 1 wherein the object is of a type,
and wherein the step of generating a customized model of the object
includes the step of: selecting a standard model of a structure of
the same type as the object, from a set of pre-defined
three-dimensional models of structures, each standard model of each
structure characterized by a plurality of modes of deformation;
generating a synthesized image of the structure using the standard
model, the synthesized image having an associated projection angle;
comparing a projection image of the object, taken at the associated
projection angle, to the synthesized image to obtain a measurement
of differences between the synthesized image and the projection
image; and modifying the standard model of the structure to
generate a customized model of the object by selectively varying at
least one mode of deformation of the standard model and repeating
the steps of generating, comparing and modifying until the
measurement of differences is within a desired threshold.
4. The method of claim 1, wherein the density map comprises an
M.times.N array of density value entries, and wherein the step of
generating a two-dimensional density map includes the steps of:
generating a synthesized projection image of the customized model
at a projection angle associated with the projection image of the
object, the projection image and the synthesized projection image
each comprising an M.times.N array of pixels; and for each entry in
the density map, storing a value associated with a relationship
between a projection image pixel and a synthesized projection image
pixel at similar locations in the respective synthesized projection
image and projection image.
5. The method of claim 4 wherein the mathematical relationship is:
G.sub.(M,N)=DXA.sub.--PI.sub.(M,N)/DRR.sub.(M,N) wherein
DXA_PI(M,N) is a first pixel intensity value of the projection
image at location (M,N), and DRR(M,N) is a second pixel intensity
value of the synthesized projection image at location (M,N).
6. A system for assessing characteristics of an object includes: a
processor; a storage medium for storing a projection image of an
object; program code stored in the storage medium and operable when
executed upon by the processor to: generate a customized a model of
the object using a generic model of the object and the projection
image of the object, the customized model comprising a plurality of
volume elements, each volume element including density information;
generate a density map using a synthesized projection image of the
customized model and a projection image of the object; and
back-project gains from the density map into the customized model
to modify the density information of each of the volume elements to
provide a density enhanced customized model of the object and
thereby transform the generic model of the object into an object
specific model; and wherein the density enhanced customized model
of the object is stored in the storage medium.
7. The system of claim 6 wherein the density map comprises an
M.times.N array of density values and wherein the program code that
is operable to generate a density map includes program code
operable to: generate a synthesized projection image of the
customized model at a projection angle associated with the
projection image of the object, the projection image and the
synthesized projection image each comprising an M.times.N array of
pixels; and for each (M,N) entry in the density map, store a value
associated with a relationship between a first image pixel at an
(M,N) location of the projection image and a second image pixel at
an (M,N) location of the synthesized projection image pixel.
8. The system of claim 7 wherein the relationship is defined by the
below equation: G.sub.(M,N)=DXA.sub.--PI.sub.(M,N)/DRR.sub.(M,N)
wherein DXA_PI(M,N) is a pixel intensity value of the projection
image at location (M,N) and DRR(M,N) is the pixel intensity value
of the synthesized projection image at location (M,N).
9. A density map, stored in a computer readable memory of a
processing system, the density map comprising an M.times.N array of
density elements derived from an M.times.N projection image of an
object and an M.times.N synthesized projection image of a generic
three dimensional model which has been customized using the
projection image, wherein the density map is used to transform the
customized, generic three dimensional model into an object specific
three dimensional model.
10. The density map of claim 9 wherein the value of a density
element at an (M,N) location of the density map is determined in
response to a first intensity value of a pixel at the (M,N)
location of the projection image of an object and a second
intensity value of a pixel at the (M,N) location of the synthesized
projection image of a generic three dimensional model which has
been customized using the projection image of the object.
11. The density map of claim 10 wherein the value of the density
element is equal to the first pixel intensity divided by the second
pixel intensity.
Description
FIELD OF THE INVENTION
[0001] The present invention relates generally to the field of
skeletal assessment and more particularly to a system and method
for generating enhanced density distributions in a 3-D model of a
structure for use in skeletal assessment.
BACKGROUND OF THE INVENTION
[0002] Currently 2-D Dual-energy X-ray Absorptiometry is the gold
standard of osteoporosis assessment. Dual-energy X-ray
absorptiometry (DXA) is a means of performing skeletal assessment
using x-ray beams of differing energy levels. A DXA system
generally includes an examination table and a C-arm at opposite
ends of which are mounted an x-ray tube and a multi-detector array.
The patient is positioned on the examination table between the
x-ray tube and the detector array of the C-arm and a fan-shaped
beam of x-rays are directed from the tube to the detector as the
tube and detector are scanned axially along the patient. The X-rays
are alternatively pulsed at high and low energy ranges. By
comparing the relative attenuation of the x-rays at the two
energies, the contributions to the attenuation due to the soft
tissue can be subtracted. When soft tissue absorption is subtracted
out, a two dimensional image of the bony structure of the patient
remains and may be used for skeletal assessment purposes.
[0003] DXA has grown to be the gold standard for osteoporosis
assessment because it provides improved skeletal assessment results
(in comparison to single energy x-rays) while minimizing patient
exposure to radiation. However one problem with DXA is that it
fails to provide important 3-D information about the volumetric
density and spatial geometry of imaged skeletal structures. Such
3-D information may be obtained using Quantitative Computed
Tomography (QCT) Densitometry. The QCT examination is performed
using a CT scanner to obtain volumetric data. One issue with QCT
lies in whether the quality of the information obtained by 3D
structural evaluation using QCT outweighs the patient's increased
exposure to radiation, the financial costs associated with QCT and
the clinical feasibility of actual QCT practice.
[0004] Attempts have been made to leverage the information provided
by DXA to obtain skeletal assessment information of similar quality
to that provided by QCT. These attempts have involved selecting a
standard 3-D model of a skeletal structure of a patient from an
atlas of structures, and customizing the 3-D model by varying its
modes of deformation until digitally reconstructed radiographs
(DDRs) (i.e., 2-D projection images) of the 3-D model can be
`matched` to co-registered DXA projection images of the patient.
With such an arrangement a customized 3-D VXA model of a skeletal
structure may be obtained using only a handful of DXA projection
images, thereby minimizing patient exposure while theoretically
providing improved quality skeletal assessment data.
[0005] However it has been determined that not all of the skeletal
health assessment information provided by a customized VXA model
meets the quality standards of QCT. Customized VXA models have
shown great accuracy for skeletal assessment measurements related
to the spatial geometry of the structure. However, measurements
which attempt to quantify the volumetric density of the skeletal
structure, and in particular the Bone Mineral Density (BMD), have
proven less reliable. It would be desirable to identify a method
for obtaining volumetric density information having accuracy
comparable to that provided by QCT yet at DXA dosing levels.
SUMMARY OF THE INVENTION
[0006] According to one aspect of the invention, a method of
generating a density enhanced model of an object includes the steps
of generating a customized model of an object using a pre-defined
set of models in combination with at least one projection image of
the object, the customized model comprising a plurality of volume
elements including density information. A density map is generated
by relating a synthesized projection image of the customized model
to a projection image of the object. Gains from the density map may
be back-projected into the customized model to provide a density
enhanced customized model of the object. The density enhanced
customized model may then be stored for later use in skeletal
assessment and/or displayed at a user interface. Because the
density map is calculated using information from the synthesized
projection image in combination with actual projection images of
the structure, it has been shown to provide spatial geometry and
volumetric density results comparable to those of QCT but with
reduced patient exposure, equipment cost and examination time.
[0007] According to another aspect of the invention, a system for
assessing density characteristics of an object includes a
processor, a storage medium for storing a projection image of an
object and program code stored in the storage medium. The program
code is operable when executed upon by the processor to: generate a
customized a model of the object using a generic model of the
object and the projection image of the object, the customized model
comprising a plurality of volume elements, each volume element
including density information, generate a density map using a
synthesized projection image of the customized model and a
projection image of the object and back-project gains from the
density map into the customized model to modify the density
information of each of the volume elements to provide a density
enhanced customized model of the object and thereby transform the
generic model of the object into an object specific model; and
wherein the density enhanced customized model of the object is
stored in the storage medium.
[0008] According to a further aspect of the invention, a density
map is stored in a computer readable memory of a processing system.
The density map comprises an M.times.N array of density elements
derived from an M.times.N projection image of an object and an
M.times.N synthesized projection image of a generic three
dimensional model which has been customized using the projection
image of the object. The density map is constructed by comparing
the image of the object with the synthesized projection image of
the three dimensional model. The density map thus constructed is
used to transform the customized, generic three dimensional model
into an object specific three dimensional model.
BRIEF DESCRIPTION OF THE FIGURES
[0009] FIG. 1 is illustrates an exemplary skeletal assessment
environment in which the present invention may be used to generate,
store and/or display density enhanced volume dual-energy x-ray
absorptiometry (E-VXA) models;
[0010] FIG. 2 is a block diagram illustrating several components of
the system of FIG. 1 which support the present invention;
[0011] FIG. 3 is a flow diagram illustrating exemplary steps that
may be performed to generate the density enhanced VXA model of the
present invention;
[0012] FIG. 4 is a functional flow diagram provided to illustrate a
method for generating a customized VXA model for use with the
present invention;
[0013] FIG. 5 is a diagram illustrating how images may be used to
generate a density map of the present invention; and
[0014] FIG. 6 is a diagram intended to illustrate back-projection
of density from the density map into the customized VXA model to
generate the density enhanced VXA model of the present
invention.
DETAILED DESCRIPTION
[0015] According to one aspect of the invention it is realized that
a customized three-dimensional model of an imaged structure may be
further transformed by applying one or more different gains to
volume elements of the model to provide a density enhanced 3-D
model of the imaged structure. Such a density-enhanced 3-D model of
the image structure has been found to have spatial geometry and
volumetric density information similar in quality to a QCT
representation of the imaged structure. The gains that are applied
to the volume elements of the 3-D model are determined by relating
density information from projection images of the customized 3-D
model to density information from one or more DXA images of the
imaged structure. The present invention therefore provides improved
skeletal assessment using 2-D images, thereby reducing patient
exposure, procedure complexity and equipment cost.
[0016] Skeletal assessment using models of the present invention
may be performed in a variety of environments using one or more
two-dimensional images obtained from any one of a variety of
imaging devices. By way of example and not limitation, FIG. 1
illustrates a representative one of the skeletal assessment
environments 10 including an imaging device 12 for capturing one or
more two-dimensional images and a workstation 14, including a
processing tower 16 and output devices/displays such as display 17
and printer 18. The imaging device 12 of FIG. 1 may be, for
example, an osteodensitometer such as the Discovery.TM.-A QDR Bone
Densitometer, provided by Hologic, Inc. of Bedford Mass. The device
12 includes a C-arm 13 with a fan beam x-ray source on one arm and
an x-ray detector array on an opposing arm. During image capture, a
patient is positioned on the examination table between the x-ray
tube and the detector array of the C-arm 13 and a fan-shaped beam
of x-rays are directed from the tube to the detector as the tube
and detector are scanned axially along the patient. The X-rays are
alternatively pulsed at high and low energy ranges and one or more
images are captured and processed to provide Dual X-ray
Absorptiometry (DXA) projection images.
[0017] It should be noted that although the specification describes
a method of constructing a density enhanced VXA model using one or
more DXA images, the present invention is not limited to use with
DXA images. DXA images are discussed because at present they are
preferred for osteoporosis assessment. However it will be
appreciated that the methods described herein may be used with any
2-D projection images of a structure, whether it be DXA, x-ray etc.
In addition it should be noted that it is not required that the 2-D
images be obtained immediately prior to skeletal assessment; rather
the present invention may be used to provide improved skeletal
assessment for images previously captured and stored, for images
captured on film and digitized, etc.
[0018] The device 12 is coupled via a network to the workstation
14. The workstation 14 includes software for controlling the scan
operation of device 12 as well as for processing and analyzing the
DXA projection images. FIG. 2 illustrates functional components of
the processing tower 16 in more detail. In general the processing
system includes one or more central processors, functional modules
and data structures 31-39. The functional modules may be comprised
of hardware, software or any combination thereof, while the data
structures may be stored in a local or remote memory. Any software
modules are operable, when executed upon by the processors, to
perform specific pre-defined functions. For the purposes of clarity
FIG. 2 illustrates the software programs as discrete functional
modules resident in memory blocks on the tower 16 although it is
readily appreciated by those of skill in the art that software
programs may be combined along different functional boundaries and
may reside in any storage accessible to the processors, whether it
is on the tower itself or located remotely. Accordingly the present
invention is not limited to embodiments delineated as shown in FIG.
2.
[0019] Representative components of a processing system of FIG. 2
that may be used in or by the present invention include a VXA model
generator 31 including an atlas 30, DXA projection images 33 and a
Fitting module 34. As will be described in more detail with regard
to FIGS. 3 and 4, the Fitting module 34 performs an iterative
process whereby a model from the VXA model generator 31 is deformed
and compared to the DXA projection images until a `best fit match`
is found between the model and the image. The resulting, `custom`
VXA model is then forwarded to density map generator 35. As will be
described in more detail with regard to FIG. 4, the density map
generator uses information from the custom 3D VXA model together
with information from the DXA projection images to generate a
density map 36. The density map 36 is forwarded together with the
custom 3D VXA model to a density application module 37 which
applies the gains from the density map to the custom 3D VXA model
to provide a density-enhanced 3D VXA model. The density-enhanced 3D
VXA model may then be forwarded to the skeletal assessment module
38, and results of skeletal assessment may be output via a user
interface to one or more displays 28 or other output devices.
[0020] FIG. 3 is a flow diagram provided to describe, in more
detail, exemplary steps that may be performed in the generation of
a density enhanced VXA model of the present invention. As mentioned
previously, DXA projection data alone fails to provide important
3-D information about the volumetric density and spatial geometry
of imaged skeletal structures. Efforts have therefore been made to
generate a three dimensional volumetric dual x-ray absorptiometry
(VXA) model using the limited number of DXA projection images. One
technique involves generating a VXA model using a deformable
density atlas. The atlas is a statistical data set that is
constructed from a large set of Computed Tomography (CT) scans
which mathematically embodies an a priori shape of a structure. The
statistical atlas includes one or more models of segmented CT data,
wherein each model may have one or more different modes of
deformation defining anatomical differences of the structure. For
example using a large number of femurs (i.e. fifty or more) a
specific tetrahedral mesh model of the femur is formed by
mathematically fitting the femur data to a tetrahedral mesh model
by deforming the average femur mesh to fit each feature of the
femur. Normal, orthogonal modes of variation for the model may then
be identified using Principal Component Analysis (PCA) or robust
PCA techniques known to those of skill in the art. Thus the atlas
comprises a set of deformable models of bone structures, each model
having defined modes of deformation which are typically orthogonal
such that modification of one mode does not affect the other
deformation modes.
[0021] As known in the art of 3-D modeling, a 3-D model is
generally represented by a set of volume elements, each volume
element defining a three dimensional structure having one or more
attributes defining characteristics of the volume element including
spatial attributes and composition attributes. For example, spatial
attributes may include information as to the location of the volume
element within the 3-D structure, while composition attributes may
provide information such as the intensity of pixels in the volume
element, where the intensity relates to the density of the
structure. The shape of the three dimensional structure of a volume
elements is a matter of design choice, and includes but is not
limited to rectilinear `voxels` or tetrahedral `tetras`. It is
noted that the various forms of volume rendering each have their
own advantages. For example, Tetrahedral Mesh Modeling of Density
Data for Anatomical Atlases and Intensity-Based Registration, by
Yao and Taylor, in MICCAI 2000, 2000. Pittsburgh Pa., USA, Springer
and Construction and Simplification of Bone Density Models, by Yao
and Taylor, Proceedings of SPIE Vo. 4322 (2001) pp. 814-823 (both
incorporated herein by reference) disclose construction of a
hierarchical tetrahedral mesh model for use in a bone density atlas
for at least the reason that tetrahedral meshes can be stored more
efficiently than voxelized representations. Thus the 3-D model may
be represented in terms a voxel or as a tetrahedral mesh.
Converting between tetrahedral models and voxelized models may
readily be accomplished by those of skill in the art.
[0022] The initial step 210 of the process of FIG. 3 involves
generating a `customized VXA model` by deforming a generic 3-D
model using information from a limited number (for example 4 or
less) of DXA projection images. FIG. 4 illustrates the process of
step 210 (also described in the Yao reference) in more detail. At
step 210(a) one or more Digitally Reconstructed Radiographs (DRR)
are synthesized and compared to associated DXA projection images.
Synthesizing DRR images involves transforming between a coordinate
system associated with the 3D model and the 2D coordinate system of
the DXA projection images. For example, the coordinate system of
the 2D DXA images may be defined in terms of projection geometry
parameters, including image center, pixel size, camera focal
length, projection angle, etc. A variety of techniques known to
those of skill in the art may be used to generate 2D images of the
3D model that are registered to coordinate system of the 2D DXA
images and the present invention is not limited to any particular
method of projection. The resulting DRR is a 2D image matched in
image and pixel size to the DXA projection image. FIG. 4
illustrates 3 projection images (211a, 211b and 211c), and 3
co-registered DRRs (212a, 212b and 212c).
[0023] At step 201(b) a comparison is performed between each of the
projected DRRs and the corresponding DXA projection images to
determine whether the 3D model `matches` the imaged structure
associated with the DXA projection images. A variety of methods of
determining when there is a `best fit match` between the imaged
structure and the 3D VXA model can be used. For example, a
pixel-by-pixel comparison of the images may be performed to
identify differences in pixel intensity value, and a model may be
determined to be a best fit match if the differences in pixel
intensity value for the image(s) are below a pre-identified
threshold. Other methods, such as the as simulated annealing and
amoeba methods of minimization known to those of skill in the art
are also within the scope of the present invention. Should the 3D
VXA model which generated the DRR images fail to be a `best fit
match` to the images structure any one of steps 210(c), 210(d) or
210(e) may be executed to transform (by varying the rotation,
translation or scale) of the model. Alternatively, or in
conjunction with any of the steps 210(c), 210(d) or 210(e), one or
more modes of the model may be modified at step 210(f). The process
of transforming and deforming the model, and generating DRRs and
comparing the DRRs to the DXA projection images continues until the
differences between the DRR images and DXA images are within `best
fit match` threshold. Once the best fit match has been determined,
the transformed and/or deformed VXA model is stored as a `custom
VXA` 225 model. It should be noted that the above process is merely
an exemplary method for customizing a 3D model of an imaged
structure. It is appreciated that other methods of customizing a 3D
model, for example that do not depend upon an atlas but use other
means, are known and implemented by those in the volume rendering
art. Therefore, although an exemplary method of generating a custom
3D model has been shown and described, the present invention is not
limited to the described process.
[0024] Referring back to FIG. 3, as discussed above the custom VXA
model generated using the process of step 210 has been found to
lack accuracy in its representation of the internal structure (i.e.
density) of the imaged structure. According to one aspect of the
invention it is realized that improved density representation may
be obtained by applying gains across the 3D model, where the gains
are represented in a density map that associates pixel intensity
values of the DXA projection images with the pixel intensity values
of the DRRs of the custom VXA model according to the below Equation
I.
G.sub.(M,N)=DXA.sub.--PI.sub.(M,N)/DRR.sub.(M,N) Equation I
Wherein the DXA projection image, DRR synthesized image and Density
Map comprise an M.times.N array of elements, and wherein
DXA_PI(M,N) is the pixel intensity value of the DXA projection
image at location (M,N) of the DXA image, and DRR(M,N) is the pixel
intensity value of the DRR at location (M,N) of the DRR.
[0025] At step 220 at least one DRR 226 is synthesized from the
customized VXA model 225 and at step 230 at least one DXA
projection image is compared to the synthesized DRR 226 to generate
density map 235. FIG. 5 is a pictorial representation of steps 220
and 230. Although FIG. 5 illustrates that multiple DRR images and
DXA images are used for density map generation, it should be
appreciated that, although possible, it is not necessary to employ
more than a single DXA/DRR image pair for density map generation
and therefore only one DXA/DRR image pair is shown with solid lines
(rather than dashed) as input to the generator 35. However
alternate embodiments where multiple density maps are initially
generated and combined, for example by averaging the resulting
gains, or by selecting a maximum or minimum gain, or some
combination thereof, are also within the scope of the present
invention.
[0026] Once the density map is generated, at step 240 it is back
projected into the custom VXA model to generate a density enhanced
custom VXA model 228. Back-projection involves identifying all of
the voxels in the 3-D structure that contribute to an intensity
value of a particular 2-D projected pixel, and applying the
computed density for that pixel to each of the voxels. For example,
referring now to FIG. 6 the density map 235 was generated from a
DRR and DXA image having a particular projection angle relative to
the imaging source 610 indicated by line 612; that the x-rays from
the imaged source that follow the trajectory of line 612 will
impact the value of the pixel X in density map 235. In the 3-D
tetrahedral mesh model, the x-ray traverses multiple tetra (or
voxels) in its path through the model. During back-projection, the
calculated density value from pixel x is applied to all of the
volume elements along the path (indicated by line 600). Regardless
of how the density of the 3-D model is represented (i.e., as a
discrete value or as a density function), the density is multiplied
to the density representation.
[0027] Because the density is calculated using patient specific DXA
projection information, the resulting density enhanced 3D VXA model
includes patient specific density information. As a result the
model is capable of being used to obtain skeletal assessment
information that is comparable to that QCT, but which uses a
limited number of x-ray views and therefore reduces patient
exposure and procedure cost.
[0028] Accordingly a method and system for generating a density
enhanced 3D model for use in skeletal assessment has been shown and
described. Having described exemplary embodiments, it can be
appreciated that the examples described above are only illustrative
and that other examples also are encompassed within the scope of
the appended claims.
* * * * *